Deprecated: The each() function is deprecated. This message will be suppressed on further calls in /home/zhenxiangba/zhenxiangba.com/public_html/phproxy-improved-master/index.php on line 456
US11537898B2 - Generative structure-property inverse computational co-design of materials - Google Patents
[go: Go Back, main page]

US11537898B2 - Generative structure-property inverse computational co-design of materials - Google Patents

Generative structure-property inverse computational co-design of materials Download PDF

Info

Publication number
US11537898B2
US11537898B2 US16/799,410 US202016799410A US11537898B2 US 11537898 B2 US11537898 B2 US 11537898B2 US 202016799410 A US202016799410 A US 202016799410A US 11537898 B2 US11537898 B2 US 11537898B2
Authority
US
United States
Prior art keywords
property
network
generator
simulated
real
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active, expires
Application number
US16/799,410
Other languages
English (en)
Other versions
US20210103822A1 (en
Inventor
Ganesh Hegde
Harsono S. Simka
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Samsung Electronics Co Ltd
Original Assignee
Samsung Electronics Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Samsung Electronics Co Ltd filed Critical Samsung Electronics Co Ltd
Priority to US16/799,410 priority Critical patent/US11537898B2/en
Priority to KR1020200063120A priority patent/KR20210040248A/ko
Assigned to SAMSUNG ELECTRONICS CO., LTD. reassignment SAMSUNG ELECTRONICS CO., LTD. ASSIGNMENT OF ASSIGNORS INTEREST (SEE DOCUMENT FOR DETAILS). Assignors: HEGDE, GANESH, SIMKA, HARSONO S.
Priority to EP20179214.0A priority patent/EP3800586B1/en
Priority to TW109130672A priority patent/TWI846942B/zh
Priority to CN202011064450.9A priority patent/CN112599208B/zh
Priority to JP2020167087A priority patent/JP7733438B2/ja
Publication of US20210103822A1 publication Critical patent/US20210103822A1/en
Publication of US11537898B2 publication Critical patent/US11537898B2/en
Application granted granted Critical
Active legal-status Critical Current
Adjusted expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/088—Non-supervised learning, e.g. competitive learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/10—Geometric CAD
    • G06F30/13—Architectural design, e.g. computer-aided architectural design [CAAD] related to design of buildings, bridges, landscapes, production plants or roads
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/20—Ensemble learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G06N3/0454—
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G06N3/0455—Auto-encoder networks; Encoder-decoder networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0475—Generative networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/094—Adversarial learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N7/00—Computing arrangements based on specific mathematical models
    • G06N7/01—Probabilistic graphical models, e.g. probabilistic networks
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
    • G16C20/00—Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
    • G16C20/50—Molecular design, e.g. of drugs
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
    • G16C20/00—Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
    • G16C20/70—Machine learning, data mining or chemometrics
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
    • G16C60/00—Computational materials science, i.e. ICT specially adapted for investigating the physical or chemical properties of materials or phenomena associated with their design, synthesis, processing, characterisation or utilisation

Definitions

  • the present disclosure relates generally to systems and methods for material design utilizing machine learning.
  • Machine learning has been utilized for material discovery to meet the demand for materials with novel and/or improved properties.
  • a physical property of interest (P) may be first identified in act 101 . This could be, for instance, adhesion of a given material to a target material, resistivity, thermal conductivity, melting point, electromigration resistance, oxidation resistance, breakdown voltage, or a combination of relevant physical properties of interest for a given application.
  • the process illustrated in FIG. 1 may be referred to as the “forward” mode of material design and discovery (e.g., “forward design”).
  • the list of candidate materials may be selected based on prior knowledge of materials suitable for property P or on the basis of chemical intuition based on the similarity of a target material to other materials known to have a suitable value of P. While known physics/algorithm that maps structure S to property P is used in forward design, such a selection process may be blind to a number of target materials that may still be suitable for a particular application but lie either (1) outside the intuition of experts in a field, or (2) are dissimilar to known materials having a given property.
  • forward design may be limited by the time required to simulate property P for structure S. Out of the possible universe of materials, only a tiny fraction can be simulated due to time constraints.
  • An aspect according to one or more embodiments of the present disclosure is directed toward a method for material design utilizing machine learning, where the underlying joint distribution p(S,P) of structure (S)-property (P) relationships is learned and is utilized to directly generate samples having (S,P) in a single step utilizing generative techniques, without additional processing steps.
  • the subspace of structures e.g., from a “whole” space representing all possible structures
  • the targets for property P are identified either utilizing conditional generation of the distribution (e.g., p(P)), or through randomly generating a large number of samples (S,P) and filtering (e.g., selecting) those that meet the target property criteria.
  • the target property (P) includes one or more physical quantities
  • the structure (S) includes a representation selected from an encoded vectorized representation in which elements of a structure vector represent atomic species in specific locations, a floating point vectorized representation in which contiguous tuples of a structure vector represent variables (x, y, z) of Cartesian coordinates of a fixed atomic species, and a mixed representation including species and/or coordinate tuples in succession.
  • the system includes a single generator network.
  • the generator network includes a first generator network to generate the simulated structure and a second generator network to generate the simulated property, and both the first generator network and the second generator network are configured to receive the random vectors from the same latent space vector generator.
  • the generator network includes a first generator network to generate the simulated structure and a plurality of second generator networks, each to generate one of a plurality of simulated properties and both the first generator network and the plurality of second generator networks are configured to receive the random vectors from the same latent space vector generator.
  • the discriminator network is configured to output a floating point or discrete count value.
  • the system includes a computing device including graphics processing units (GPUs) and/or field programmable gate arrays (FPGAs).
  • GPUs graphics processing units
  • FPGAs field programmable gate arrays
  • a method to generate a structure (S) for a target property (P) utilizing machine learning includes: training a discriminator network with a database including real structures and real target property, generating a simulated structure and a simulated property through a generator network, training the generator network so that the discriminator network fails to distinguish a difference between the simulated structure and the real structure from the database and learning an underlying joint probability distribution of structure-property p(S, P) simultaneously, and generating the structure for the target property utilizing the underlying joint probability distribution.
  • the generating of the structure for the target property includes utilizing latent space partitioning p(P) to generate the structure (S) for the target property (P).
  • the generating of the structure for the target property includes randomly generating a large number of samples (S,P) and filtering those that meet the target property (P).
  • the target property (P) includes a physical quantity
  • the structure (S) includes a representation selected from an encoded vectorized representation in which elements of a structure vector represent atomic species in specific locations, a floating point vectorized representation in which contiguous tuples of a structure vector represent variables (x, y, z) of Cartesian coordinates of a fixed atomic species, and a mixed representation containing (species, coordinates) tuples in succession.
  • the method further includes generating random vectors through a latent space vector generator, wherein the generating of the simulated structure and the generating of the simulated property through the generator network utilizes the random vectors as inputs, and wherein the training of the generator network includes sending a loss function from the discriminator network to the generator network.
  • the generator network includes a first generator network to generate the simulated structure and a second generator network to generate the simulated property, and both the first generator network and the second generator network are configured to receive the random vectors from the same latent space vector generator.
  • a machine learning system to generate a structure (S) for a target property (P) includes a variational autoencoder (VAE), wherein the VAE includes an encoder network, a latent space, and a decoder network, wherein the encoder network is configured to receive a database including real structures and real properties as an input and compress the input to generate encoded vectors, mean and standard deviation of a distribution of the encoded vectors in the latent space, the decoder network is configured to be trained from the database including real structures and real properties, to decompress the encoded vectors to create simulated structure and simulated property, and to distinguish a difference between the simulated structure and the simulated property and the real structure and the real property, and to send a loss function to the encoder network, wherein the encoder network, the latent space, and the decoder network are arranged to learn an underlying joint probability distribution of structure-property p(S, P) simultaneously.
  • VAE variational autoencoder
  • the target property (P) includes one or more physical quantities
  • the structure (S) includes a representation selected from an encoded vectorized representation in which elements of a structure vector represent atomic species in specific locations, a floating point vectorized representation in which contiguous tuples of a structure vector represent variables (x, y, z) of Cartesian coordinates of a fixed atomic species, and a mixed representation including species and/or coordinate tuples in succession.
  • the system includes a computing device including graphics processing units (GPUs) and/or field programmable gate arrays (FPGAs).
  • GPUs graphics processing units
  • FPGAs field programmable gate arrays
  • a method to generate a structure (S) for a target property (P) utilizing machine learning includes: inputting a database including real structures and real properties to an encoder network as an input and compressing the input to generate encoded vectors, mean and standard deviation of a distribution of the encoded vectors in a latent space, training a decoder network with the database including real structures and real properties, decompressing the encoded vectors to create simulated structure and simulated property by the decoder network, sending a loss function to the encoder network, and learning an underlying joint probability distribution of structure-property p(S, P) simultaneously.
  • the method further includes generating the structure for the target property utilizing latent space partitioning p(P) to generate the structure (S) for the target property (P).
  • the method further includes generating the structure for the target property through randomly generating a large number of samples (S,P) and filtering those that meet the target property (P).
  • the target property (P) includes a physical quantity
  • the structure (S) includes a representation selected from an encoded vectorized representation in which elements of a structure vector represent atomic species in specific locations, a floating point vectorized representation in which contiguous tuples of a structure vector represent variables (x, y, z) of Cartesian coordinates of a fixed atomic species, and a mixed representation containing (species, coordinates) tuples in succession.
  • FIG. 1 is an illustration of a forward design process.
  • FIG. 2 is a schematic illustration of a system and a training process for a generative adversarial network (GAN) according to an embodiment of the present disclosure.
  • GAN generative adversarial network
  • FIG. 3 is schematic illustration of a sample generation process according to an embodiment of the present disclosure.
  • FIG. 4 is a schematic illustration of a system and a training process for a generative adversarial network (GAN) according to an embodiment of the present disclosure.
  • GAN generative adversarial network
  • FIG. 5 is a schematic illustration of inverse design approaches for crystal structure prediction.
  • FIG. 6 is a schematic representation of inverse design approaches for a metasurface.
  • FIG. 7 is a schematic a system and a training process utilizing a VAE architecture according to an embodiment of the present disclosure.
  • the electronic or electric devices and/or any other relevant devices or components according to embodiments of the present invention described herein may be implemented utilizing any suitable hardware, firmware (e.g. an application-specific integrated circuit), software, or a combination of software, firmware, and hardware.
  • the various components of these devices may be a process or thread, running on one or more processors, in one or more computing devices, executing computer program instructions and interacting with other system components for performing the various functionalities described herein.
  • the computer program instructions are stored in a memory which may be implemented in a computing device utilizing a standard memory device, such as, for example, a random access memory (RAM).
  • the computer program instructions may also be stored in other non-transitory computer readable media such as, for example, a CD-ROM, flash drive, or the like.
  • structure may refer to atomic structure, arrangement of atoms in regular or irregular patterns in materials, and/or the spatial composition of atomic species in a material (e.g., a compound).
  • the term “property” may refer to a physical property (such as boiling point, melting point, specific heat capacity, density, resistivity, thermal conductivity, electronic transmission, electromigration resistance, density of electronic levels, density of phononic levels, the mean free path of electron-phonon scattering, etc.); and/or an electronic structure property computed based on the atomic arrangement defined in the structure.
  • a physical property such as boiling point, melting point, specific heat capacity, density, resistivity, thermal conductivity, electronic transmission, electromigration resistance, density of electronic levels, density of phononic levels, the mean free path of electron-phonon scattering, etc.
  • joint probability distribution p(S, P) refers to the probability distribution representing the structure and property relationships, where both the structure S and the target property P, as two separate events, would be found simultaneously.
  • the expression “the joint distribution p(S,P) of structure (S)-property (P) relationship is learned simultaneously” refers to a system or method that learns the relationship p(S,P) directly, simultaneously or concurrently.
  • the joint distribution p(S,P) of structure (S)-property (P) relationship is learned simultaneously,” the joint distribution p(S,P) is not obtained from the individual probability p(S), the individual probability p(P), the conditional probability p(S
  • sample (S, P) refers to a sample with the structure S and property P.
  • a method and a system for material design utilizing machine learning are provided, where the underlying joint probability distribution p(S,P) of structure (S)-property (P) relationships is learned simultaneously (i.e., explicitly learned simultaneously) and is utilized to directly generate samples (S,P) in a single step (act) utilizing generative techniques, without any additional processing steps (acts).
  • the subspace of structures that meet or exceed the target for property P is then identified utilizing conditional generation of the distribution (e.g., p(P)), or through randomly generating a large number of samples (S,P) and filtering (e.g., selecting) those that meet target property criteria.
  • an alternative approach to forward design may be materials inverse design.
  • the target property P may be mapped on to structure S by utilizing a number of inverse design approaches.
  • inverse design approaches may be utilized for crystal structure prediction as shown in the schematic illustration of FIG. 5 .
  • a statistical inference-based approach 510 entails learning structure-property relationships from a large database and utilizes this learning to predict the probability of a certain composition of materials to have a particular property, e.g., a particular crystal structure type.
  • the output is the crystal (material) composition
  • the input is the crystal structure.
  • the composition versus ground state crystal structure mapping is extracted into a machine learning model. The model is then utilized to predict the crystal type for a composition having an unknown structure.
  • the property ⁇ or the sum over a number of properties is computed in a space of candidate structures ⁇ .
  • a global minimum i.e., the structure that optimizes the difference between desired properties and computed properties is obtained by utilizing heuristic optimization techniques.
  • Another approach may be a generative model-based approach 550 .
  • the probability distribution of interest is p(property
  • the category of problems where the distribution of the outputs is sought when the input is known are called discriminative problems.
  • the conditionality is reversed—p(structure
  • GAN generative adversarial networks
  • metasurface design may be transitioned from trial-and-error approaches to neural network mediated inverse design by utilizing both simulation and inverse design to enact structure-property relationships to generate an optical spectrum from a metasurface and vice-versa. Both processes may be replaced by deep neural networks.
  • the architecture of a network for artificial intelligence (AI)-based optical design may include three networks: the generator 610 , the simulator 630 , and the critic 650 as shown in FIG. 6 .
  • the generator may accept the spectra T and noise z and produce possible patterns.
  • the simulator may be a pre-trained network that approximates the transmittance spectrum T ⁇ circumflex over ( ) ⁇ for a given pattern at its output, and the critic may evaluate the distance of the distributions between the geometric data and the patterns from the generator. While training the generator, the produced patterns may vary according to the feedback obtained from the simulator and critic. Valid patterns may be documented during the training process, and may be smoothed to qualify as candidate structures, but additional steps (acts) will be required to identify the structure for a given property.
  • the approach according to embodiments of the present disclosure may perform inverse design entirely in a generative framework without requiring any additional learning framework to bias the distribution towards desired/target properties.
  • reinforcement learning may not be needed (i.e., may not be included) in the present application (e.g., current method).
  • the joint probability distribution of structures and properties of arbitrary materials can be obtained in a single step (act) instead of including intermediate steps (acts) and post processing.
  • (S, P), i.e., structure and property may be produced in a single step (act) without any intermediate or post processing steps (acts).
  • a method for generating structure and property simultaneously includes learning the underlying statistical properties of the joint probability distribution of structure-property p(S, P) simultaneously, where no additional steps (acts) or computational processing are needed between the generation of the structure and the generation of the property.
  • the joint distribution p(S,P) is not learned sequentially by learning the distribution of p(S), p(S
  • the underlying joint probability distribution p(S,P) may be learned (or estimated) by a generative framework selected from, but not limited to, generative adversarial networks (GANs), any suitable variants thereof (such as deep convolutional GAN (DCGAN), Wasserstein GAN (WGAN), Fisher GAN (FGAN), etc.), variational autoencoders (VAE), any suitable variants thereof (such as Attention network), and/or combinations thereof (such as VAE-GAN).
  • GANs generative adversarial networks
  • DCGAN deep convolutional GAN
  • WGAN Wasserstein GAN
  • Fisher GAN FGAN
  • VAE variational autoencoders
  • VAE any suitable variants thereof (such as Attention network), and/or combinations thereof (such as VAE-GAN).
  • FIG. 2 is a schematic illustration of a system and a training process for a generative adversarial network (GAN) according to an embodiment of the present disclosure.
  • the system includes a latent space vector generator (V) 210 , a generator network (G) 230 , and a discriminator network (D) 250 .
  • V latent space vector generator
  • G generator network
  • D discriminator network
  • an epoch is one iteration through all the training data.
  • the training data is split into batches of various sizes.
  • An initial pass in every training epoch teaches the discriminator what a “true” example looks like utilizing the real structure and real property pair.
  • Subsequent passes in each training epoch are utilized to train the generator network (G) to produce simulated structure and simulated property that are indistinguishable from the real pair.
  • G generator network
  • the discriminator network (D) is trained utilizing both real structures (R real ) and real properties (T real ) simultaneously to learn a mapping function between the two. For example, actual materials with the target properties are utilized to train the discriminator network (D).
  • the discriminator network (D) By training the discriminator network (D) with both real structures (R real ) and real properties (T real ), the structure-property relation can be learned simultaneously (e.g., not sequentially after learning the structure before or after learning the structure separately).
  • latent space refers to an abstract multi-dimensional space containing feature values (i.e., vectors) that encodes a meaningful internal representation of externally observed events (such as the structure and the property of materials).
  • Latent space aims to provide an understanding (or representation) of these events (structures and properties) to a computer through a quantitative spatial representation/modeling.
  • Each attribute of the event is represented by one or more vectors in the latent space, and an event may be represented with a plurality of dimensions in the latent space.
  • the latent space vector generator (V) generates purely random numbers of the desired dimensionality.
  • the latent space vector generator (V) may output a purely random vector of any suitable dimensionality, e.g., a dimensionality of 10, such that all random numbers are drawn from a Gaussian distribution with zero mean and unit variance, i.e., the standard normal distribution.
  • the idea behind utilizing a random number generator in this framework is that the meaning to these random vectors gets assigned “a posteriori”, i.e., after the discriminator network (D) rejects samples for being fake (i.e., simulated).
  • the successful completion of the training procedure finds the subspace that corresponds to the distribution of structure-property pair with materials having the desired property.
  • the latent space vector generator (V) generates random vectors that are mapped to the property space and structure space
  • the generator network (G) generates the simulated samples of (T sim , R sim ) tuples including the simulated property T sim and the simulated structure R sim based on the random vectors provided from the latent space vector generator (V).
  • T and P are utilized interchangeably to refer to the property
  • R and S are utilized interchangeably to refer to the material structure.
  • the training objective is to increase the error rate of the discriminative network (D) (i.e., to “fool” the discriminator network (D) so that it cannot distinguish the sample generated by the generator network (G) from the real structure-property pair).
  • the loss function may be any suitable type corresponding to the quantity being learned. For instance, if a continuous variable is being learned, the loss function may be a quantity such as root mean squared error. If a categorical (e.g., qualitative) variable is being learned, the loss may be cross-entropy, Kullback-Leibler divergence or similar measures. In one embodiment, a negative binomial loss function may be utilized for training the generator network (G) when the output data is discrete. However, embodiments of the present disclosure are not limited thereto, and other suitable loss functions may be utilized.
  • the structure and/or property may include or consist of ordinal or count data.
  • the discriminator network (D) may output a floating point (e.g., continuous floating point) or discrete count value instead of binary (True/False, 0/1) values.
  • the loss function may be calculated based on the output of the discriminator network (D). For example, the loss function for count data may be defined utilizing Negative Binomial loss or Poisson loss.
  • (T, R) samples that satisfy the target property P can be created by sampling the latent space through the generator network (G) as shown in FIG. 3 .
  • the latent space vector generator (V) 210 produces random vectors for the generator network (G) 230 , and the generator network (G) 210 transforms the random vectors to (S,P), i.e., a structure (S) that has the target property (P).
  • the subspace of structures that meets or exceeds the targets for property P may be identified either through use of latent space partitioning (p(P)), or through simultaneous (or concurrently) generation of a large number of candidate structure/property pairs (S,P) followed by filtering to select candidates with the desired properties (P).
  • p(P) latent space partitioning
  • S,P candidate structure/property pairs
  • the target property may be any combination of two or more desirable physical quantities, such as resistivity, density of states, etc.
  • the structure may be an encoded vectorized representation in which elements of the structure vector represent atomic species in specific locations.
  • the target property may be any combination of two or more desirable physical quantities, such as resistivity, density of states, etc.
  • the structure may be a floating point vectorized representation in which contiguous tuples of the structure vector represent variables (x, y, z) in Cartesian coordinates of a fixed atomic species.
  • the system and method according to some example embodiments may be applicable to design any desirable materials, such as 1D, 2D, 3D, molecules, etc.
  • the target property may be any combination of two or more desirable physical quantities, such as resistivity, density of states, etc.
  • the structure may be a mixed representation containing (species, coordinates) tuples in succession.
  • the generative framework for structure and properties utilizes a single generative network (G) to generate both structure and property.
  • Both structure and property can be vectors that are concatenated to each other, and then fed to the discriminator network (D).
  • FIG. 4 is a schematic illustration of a system and a training process for a generative adversarial network (GAN) according to an embodiment of the present disclosure.
  • a latent space vector generator (V) 410 feeds into two generator networks (G) 431 and 433 that separately create structure (R) and property (T) vectors respectively, that are concatenated and fed to the discriminator network (D) 450 .
  • the loss from the discriminator network (D) is fed into both generator networks (G) during the training process.
  • the loss function may be Mean Squared Error (MSE), cross entropy, or any other suitable loss functions.
  • MSE Mean Squared Error
  • cross entropy or any other suitable loss functions.
  • CrossEntropyLoss ⁇ ( y i log( ⁇ i )+(1 ⁇ y i )log(1 ⁇ i ))
  • a variational autoencoder includes an encoder network, a latent space, a decoder network, and a loss function.
  • the encoder network is a neural network that outputs a compressed representation z of data x.
  • the decoder network is a neural network that learns to reconstruct the data x given the compressed representation z. Through training, the loss function gets reduced or minimized and the VAE learns a probability distribution of the data.
  • the encoder network is composed of Convolutional Layers, which compress the input and outputs a dense representation of the input, called encodings.
  • This process where the encoder network compresses the data is referred to as data compression or dimensionality reduction.
  • the input is encoded as vectors with a probability distribution over the latent space.
  • a point from the latent space is sampled from that distribution.
  • the decoder network uses Deconvolutional Layers, which is the reverse of Convolutional Layers, and decompress the sampled point.
  • the real structure and real property pair (i.e., the initial data) is fed to the encoder network. That is, both the real structure and the real properties are fed to the encoder network.
  • the encoder network transforms the input as a vector distribution over the latent space, and learns the mean and standard deviation of the probability distribution. Then the decoder network takes the output of the encoder, and outputs a restructured version of the input.
  • the difference between the restructured version of the input and the actual input is backpropagated as the loss function to adjust the weights of the neural networks of the encoder and decoder in an iterative optimization process so that at the end of the training process, the output of the decoder network is an acceptable approximation of or can't be distinguished from the initial data.
  • the loss function is composed of a “reconstruction term” (on the final layer), that tends to make the encoding-decoding scheme as performant as possible, and a “regularization term” (on the latent layer), that tends to regularize the organization of the latent space by making the distributions returned by the encoder close to a standard normal distribution. Through the training process, the loss function is minimized.
  • the VAE architecture can be utilized to obtain new structure-property pairs.
  • the latent space becomes an encoded representation (of the structure-property pair), which when passed through a suitable decoder produces the desired output.
  • the learned joint distribution probability, mean and standard deviation are utilized to provide inputs from the latent space to be decoded by the decoder to generate (partake) new structure-property pairs that have the structures suitable to provide the target properties.
  • a filter may be utilized to further select the material structures.
  • the decoder would act similarly to the generator network of a Generative Adversarial Network and the encoder is not utilized.
  • FIG. 7 is a schematic illustration of a VAE architecture.
  • an input is transformed by the encoder network 710 to a latent distribution (as an encoded vector) in the latent space 730 .
  • the encoded vector from the latent distribution is sampled by the decoder network 750 and transformed to a decoded output through the decoding process.
  • a point from the latent space (as a sampled vector) 730 is selected and decoded by the decoder network 750 to produce a newly generated structure/property pair.
  • the generative process may incorporate uncertainty estimation, which gives a measure of how confident the model is about its predictions.
  • the uncertainty of predictions can be estimated utilizing any suitable techniques available, such as active learning, Bayesian learning, etc.
  • the property of interest may be electronic transmission (T) while the structure (R) may be represented by categorical values representing atomic species at specific positions in a unit cell.
  • T electronic transmission
  • R structure
  • categorical values representing atomic species at specific positions in a unit cell.
  • alloys such as a binary, ternary or quaternary alloys.
  • Both T and R can be vectors that are concatenated to each other.
  • T and R are discrete. For example, T can take integer values while R can take categorical values.
  • the method can be applied to the joint probability distribution of structure to multiple properties, e.g., structure ⁇ (property 1 , property 2 ,), relationships.
  • the first property may be the electronic transmission and the second property may be density of electronic levels.
  • the method can be made conditional on certain properties. For example, the method may be utilized to find p(S, P 1
  • the structure obtained from methods according to embodiments of the present disclosure are not specifically limited to molecular structures, and may further include atomic structure, arrangement of atoms in regular or irregular patterns in materials, and the spatial composition of atomic species in a material (e.g., compound) or a sample of a material.
  • the system may be implemented utilizing any suitable computing devices, such as those with graphics processing units (GPUs), field programmable gate arrays (FPGAs), or other suitable hardware.
  • GPUs graphics processing units
  • FPGAs field programmable gate arrays

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Medical Informatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Geometry (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Hardware Design (AREA)
  • Chemical & Material Sciences (AREA)
  • Pure & Applied Mathematics (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Probability & Statistics with Applications (AREA)
  • Crystallography & Structural Chemistry (AREA)
  • Structural Engineering (AREA)
  • Civil Engineering (AREA)
  • Architecture (AREA)
  • Databases & Information Systems (AREA)
  • Medicinal Chemistry (AREA)
  • Pharmacology & Pharmacy (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Algebra (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
US16/799,410 2019-10-02 2020-02-24 Generative structure-property inverse computational co-design of materials Active 2040-12-02 US11537898B2 (en)

Priority Applications (6)

Application Number Priority Date Filing Date Title
US16/799,410 US11537898B2 (en) 2019-10-02 2020-02-24 Generative structure-property inverse computational co-design of materials
KR1020200063120A KR20210040248A (ko) 2019-10-02 2020-05-26 물질의 생성 구조-특성 역 계산 공동 설계
EP20179214.0A EP3800586B1 (en) 2019-10-02 2020-06-10 Generative structure-property inverse computational co-design of materials
TW109130672A TWI846942B (zh) 2019-10-02 2020-09-08 用以針對目標性質生成結構的機器學習系統以及方法
CN202011064450.9A CN112599208B (zh) 2019-10-02 2020-09-30 生成目标材料属性的材料结构的机器学习系统和方法
JP2020167087A JP7733438B2 (ja) 2019-10-02 2020-10-01 マシンラーニングシステム及び方法

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US201962909648P 2019-10-02 2019-10-02
US16/799,410 US11537898B2 (en) 2019-10-02 2020-02-24 Generative structure-property inverse computational co-design of materials

Publications (2)

Publication Number Publication Date
US20210103822A1 US20210103822A1 (en) 2021-04-08
US11537898B2 true US11537898B2 (en) 2022-12-27

Family

ID=71092247

Family Applications (1)

Application Number Title Priority Date Filing Date
US16/799,410 Active 2040-12-02 US11537898B2 (en) 2019-10-02 2020-02-24 Generative structure-property inverse computational co-design of materials

Country Status (6)

Country Link
US (1) US11537898B2 (ja)
EP (1) EP3800586B1 (ja)
JP (1) JP7733438B2 (ja)
KR (1) KR20210040248A (ja)
CN (1) CN112599208B (ja)
TW (1) TWI846942B (ja)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11837333B1 (en) 2022-12-20 2023-12-05 Dow Global Technologies Llc Simulation guided inverse design for material formulations
US20230409781A1 (en) * 2022-06-15 2023-12-21 Korea University Research & Business Foundation Inverse design system designing nano-optical device using controllable generative adversarial network and training and design methods
US12368503B2 (en) 2023-12-27 2025-07-22 Quantum Generative Materials Llc Intent-based satellite transmit management based on preexisting historical location and machine learning
US12587274B2 (en) 2023-03-28 2026-03-24 Quantum Generative Materials Llc Satellite optimization management system based on natural language input and artificial intelligence
US12603701B2 (en) 2023-12-27 2026-04-14 Quantum Generative Materials Llc Distributed satellite constellation management and control system

Families Citing this family (22)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
GB201718756D0 (en) * 2017-11-13 2017-12-27 Cambridge Bio-Augmentation Systems Ltd Neural interface
US11196503B2 (en) * 2020-02-20 2021-12-07 X Development Llc Techniques of robust inverse design that account for manufacturing variabilities due to operating conditions
US12135927B2 (en) 2020-03-31 2024-11-05 International Business Machines Corporation Expert-in-the-loop AI for materials discovery
US12347530B2 (en) * 2020-03-31 2025-07-01 International Business Machines Corporation Expert-in-the-loop AI for materials generation
EP3958182B1 (en) * 2020-08-20 2025-07-30 Dassault Systèmes Variational auto-encoder for outputting a 3d model
US11403496B2 (en) 2020-11-04 2022-08-02 Silicon Laboratories Inc. Detecting anomalous events in a discriminator of an embedded device
US20240194305A1 (en) * 2021-04-16 2024-06-13 The Regents Of The University Of California Machine learning enabled techniques for material design and ultra-incompressible ternary compounds derived therewith
KR102370993B1 (ko) * 2021-05-25 2022-03-07 주식회사 디블렌트 신경망 기반의 실시간 수어 통역 및 대화를 지원하는 인공지능 시스템
CN114329900B (zh) * 2021-11-30 2025-06-27 浪潮电子信息产业股份有限公司 一种基于神经网络的器件参数获取方法、系统及相关组件
CN114861520A (zh) * 2022-03-24 2022-08-05 中国科学院上海微系统与信息技术研究所 一种宽频光学器件的逆设计方法
EP4394780A1 (en) * 2022-12-27 2024-07-03 Basf Se Methods and apparatuses for generating a digital representation of chemical substances, measuring physicochemical properties and generating control data for synthesizing chemical substances
KR20250006884A (ko) * 2022-04-14 2025-01-13 바스프 에스이 화학 물질들을 특성화하고, 물리화학 속성들을 측정하고, 화학 물질들을 합성하기 위한 제어 데이터를 생성하기 위한 방법들 및 장치들
DE102022204087A1 (de) 2022-04-27 2023-11-02 Robert Bosch Gesellschaft mit beschränkter Haftung Computerimplementiertes Verfahren zum Design einer synthetischen Mikrostruktur für einen polykristallinen Werkstoff, Verfahren zur Herstellung eines polykristallinen Werkstoffs und Vorrichtung dafür
US20240062068A1 (en) * 2022-08-17 2024-02-22 Accenture Global Solutions Limited Mixed synthetic data generation
WO2024085414A1 (ko) * 2022-10-17 2024-04-25 삼성전자주식회사 전자 장치 및 그 제어 방법
CN115691695B (zh) * 2022-11-11 2023-06-30 中南大学 基于gan和vae的材料成分生成方法及评价方法
CN115796037B (zh) * 2022-12-02 2026-03-06 中南大学 基于rvae的三元正极材料制备过程监测方法及系统
CN117766070B (zh) * 2023-06-08 2024-12-17 香港量子人工智能实验室有限公司 一种基于人工智能的材料设计方法、装置、终端及介质
CN116913436B (zh) * 2023-08-10 2024-04-05 华东交通大学 一种基于ldm-pnn和粒子群算法的超原子逆向设计方法
CN119669707B (zh) * 2025-02-20 2025-04-22 集美大学 一种基于深度生成模型的钙钛矿组分比例分析方法
CN120408866A (zh) * 2025-07-02 2025-08-01 数格致元(天津)科技有限公司 基于人工智能的三维参数化模型的生成方法及装置
CN121413459A (zh) * 2025-12-26 2026-01-27 国网上海市电力公司 一种基于多模态深度学习与逆向设计的超导磁体优化方法

Citations (23)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6763309B2 (en) 2001-08-06 2004-07-13 Novodynamics, Inc. Method and system for the development of materials
US7292958B2 (en) 2004-09-22 2007-11-06 Massachusetts Institute Of Technology Systems and methods for predicting materials properties
US7882052B2 (en) 2006-09-11 2011-02-01 Eors Szathmary Evolutionary neural network and method of generating an evolutionary neural network
US8671369B2 (en) 2009-12-08 2014-03-11 University Of Seoul Industry Cooperation Foundation Quantum Karnaugh map
US20170039469A1 (en) 2015-08-04 2017-02-09 Qualcomm Incorporated Detection of unknown classes and initialization of classifiers for unknown classes
WO2017094899A1 (en) 2015-12-02 2017-06-08 Preferred Networks, Inc. Generative machine learning systems for drug design
US20180032663A1 (en) 2016-08-01 2018-02-01 Samsung Electronics Co., Ltd. Method and apparatus for searching new material
US20180120479A1 (en) 2016-10-31 2018-05-03 The Government Of The United States Of America, As Represented By The Secretary Of The Navy Porosity prediction based on effective stress
US10043261B2 (en) 2016-01-11 2018-08-07 Kla-Tencor Corp. Generating simulated output for a specimen
US20180225426A1 (en) 2008-06-23 2018-08-09 Atonarp Inc. System for handling information relating to chemical substances
CN108596994A (zh) 2018-04-24 2018-09-28 朱高杰 一种基于深度学习和数据自洽的磁共振弥散加权成像方法
WO2018220368A1 (en) 2017-05-30 2018-12-06 Gtn Ltd Tensor network machine learning system
US20190155977A1 (en) 2017-11-17 2019-05-23 Samsung Electronics Co., Ltd. Method of screening materials using forward conducting modes
US10360506B2 (en) 2014-07-23 2019-07-23 Hrl Laboratories, Llc General formal concept analysis (FCA) framework for classification
US20190251459A1 (en) 2017-08-28 2019-08-15 Korea Institute Of Science And Technology Method and electronic device for predicting electronic structure of material
US20190302290A1 (en) * 2018-03-27 2019-10-03 Westerngeco Llc Generative adversarial network seismic data processor
US20190354801A1 (en) * 2018-05-16 2019-11-21 Nec Laboratories America, Inc. Unsupervised cross-domain distance metric adaptation with feature transfer network
US20190354804A1 (en) * 2018-05-15 2019-11-21 Toyota Research Institute, Inc. Systems and methods for conditional image translation
CN111144565A (zh) 2019-12-27 2020-05-12 中国人民解放军军事科学院国防科技创新研究院 基于一致性训练的自监督领域自适应深度学习方法
US20200201905A1 (en) 2018-12-20 2020-06-25 International Business Machines Corporation Methods of automatically and self-consistently correcting genome databases
US20200363485A1 (en) 2019-04-08 2020-11-19 University Of Virginia Patent Foundation Multiband spiral cardiac mri with non-cartesian reconstruction methods
US20200394275A1 (en) * 2019-06-14 2020-12-17 Palo Alto Research Center Incorporated Design of microstructures using generative adversarial networks
US20210304847A1 (en) * 2018-09-21 2021-09-30 Deepmind Technologies Limited Machine learning for determining protein structures

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11580375B2 (en) * 2015-12-31 2023-02-14 Kla-Tencor Corp. Accelerated training of a machine learning based model for semiconductor applications
US9971958B2 (en) * 2016-06-01 2018-05-15 Mitsubishi Electric Research Laboratories, Inc. Method and system for generating multimodal digital images
JP7047115B2 (ja) * 2018-02-17 2022-04-04 リジェネロン・ファーマシューティカルズ・インコーポレイテッド Ｍｈｃペプチド結合予測のためのｇａｎ－ｃｎｎ

Patent Citations (24)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6763309B2 (en) 2001-08-06 2004-07-13 Novodynamics, Inc. Method and system for the development of materials
US7292958B2 (en) 2004-09-22 2007-11-06 Massachusetts Institute Of Technology Systems and methods for predicting materials properties
US7882052B2 (en) 2006-09-11 2011-02-01 Eors Szathmary Evolutionary neural network and method of generating an evolutionary neural network
US20180225426A1 (en) 2008-06-23 2018-08-09 Atonarp Inc. System for handling information relating to chemical substances
US8671369B2 (en) 2009-12-08 2014-03-11 University Of Seoul Industry Cooperation Foundation Quantum Karnaugh map
US10360506B2 (en) 2014-07-23 2019-07-23 Hrl Laboratories, Llc General formal concept analysis (FCA) framework for classification
US20170039469A1 (en) 2015-08-04 2017-02-09 Qualcomm Incorporated Detection of unknown classes and initialization of classifiers for unknown classes
WO2017094899A1 (en) 2015-12-02 2017-06-08 Preferred Networks, Inc. Generative machine learning systems for drug design
US20170161635A1 (en) 2015-12-02 2017-06-08 Preferred Networks, Inc. Generative machine learning systems for drug design
US10043261B2 (en) 2016-01-11 2018-08-07 Kla-Tencor Corp. Generating simulated output for a specimen
US20180032663A1 (en) 2016-08-01 2018-02-01 Samsung Electronics Co., Ltd. Method and apparatus for searching new material
US20180120479A1 (en) 2016-10-31 2018-05-03 The Government Of The United States Of America, As Represented By The Secretary Of The Navy Porosity prediction based on effective stress
WO2018220368A1 (en) 2017-05-30 2018-12-06 Gtn Ltd Tensor network machine learning system
US20190251459A1 (en) 2017-08-28 2019-08-15 Korea Institute Of Science And Technology Method and electronic device for predicting electronic structure of material
US20190155977A1 (en) 2017-11-17 2019-05-23 Samsung Electronics Co., Ltd. Method of screening materials using forward conducting modes
US20190302290A1 (en) * 2018-03-27 2019-10-03 Westerngeco Llc Generative adversarial network seismic data processor
CN108596994A (zh) 2018-04-24 2018-09-28 朱高杰 一种基于深度学习和数据自洽的磁共振弥散加权成像方法
US20190354804A1 (en) * 2018-05-15 2019-11-21 Toyota Research Institute, Inc. Systems and methods for conditional image translation
US20190354801A1 (en) * 2018-05-16 2019-11-21 Nec Laboratories America, Inc. Unsupervised cross-domain distance metric adaptation with feature transfer network
US20210304847A1 (en) * 2018-09-21 2021-09-30 Deepmind Technologies Limited Machine learning for determining protein structures
US20200201905A1 (en) 2018-12-20 2020-06-25 International Business Machines Corporation Methods of automatically and self-consistently correcting genome databases
US20200363485A1 (en) 2019-04-08 2020-11-19 University Of Virginia Patent Foundation Multiband spiral cardiac mri with non-cartesian reconstruction methods
US20200394275A1 (en) * 2019-06-14 2020-12-17 Palo Alto Research Center Incorporated Design of microstructures using generative adversarial networks
CN111144565A (zh) 2019-12-27 2020-05-12 中国人民解放军军事科学院国防科技创新研究院 基于一致性训练的自监督领域自适应深度学习方法

Non-Patent Citations (23)

* Cited by examiner, † Cited by third party
Title
AKSHATKUMAR NIGAM; PASCAL FRIEDERICH; MARIO KRENN; AL\'AN ASPURU-GUZIK: "Augmenting Genetic Algorithms with Deep Neural Networks for Exploring the Chemical Space", ARXIV.ORG, CORNELL UNIVERSITY LIBRARY, 201 OLIN LIBRARY CORNELL UNIVERSITY ITHACA, NY 14853, 25 September 2019 (2019-09-25), 201 Olin Library Cornell University Ithaca, NY 14853 , XP081579685
Artur Kadurin, et al., druGAN: An Advanced Generative Adversarial Autoencoder Model for de Novo Generation of New Molecules with Desired Molecular Properties in Silico, Molecular Pharmaceutics, 2017, pp. 3098-3104, vol. 14, ACS Publishing, U.S.
Benjamin Sanchez-Lengeling, et al., Inverse molecular design using machine learning: Generative models for matter engineering, Science, 2018, pp. 360-365, vol. 361, The American Association for the Advancement of Science, U.S.
Benjamin Sanchez-Lengeling, et al., Optimizing distributions over molecular space. An Objective-Reinforced Generative Adversarial Network for Inverse-design Chemistry (ORGANIC), 2017, pp. 1-18, ChemRxiv, U.S. Website: https://chemrxiv.org/articles/ORGANIC_1_pdf/5309668/3.
Blaschke, Thomas, et al., "Application of generative autoencoder in de novo molecular design," arxiv.org, Cornell University Library, Nov. 2017, XP080839053, 13 pages.
Dong, Yuan, et al., "Inverse Structural Design of Graphene/Boron Nitride Hybrids by Regressional GAN," arxiv.org, Cornell University Library, Aug. 2019, XP081466951, 31 pages.
EPO Extended Search Report dated Dec. 11, 2020, issued in corresponding European Patent Application No. 20179214.0 (12 pages).
Geun Ho Gu, et al., Machine learning for renewable energy materials, Journal of Materials Chemistry A, Aug. 7, 2019, pp. 17096-17117, vol. 7, No. 29, The Royal Society of Chemistry, U.K.
JONATHAN SCHMIDT, MáRIO R. G. MARQUES, SILVANA BOTTI, MIGUEL A. L. MARQUES: "Recent advances and applications of machine learning in solid-state materials science", NPJ COMPUTATIONAL MATERIALS, vol. 5, no. 1, 1 December 2019 (2019-12-01), XP055733799, DOI: 10.1038/s41524-019-0221-0
Keith T. Butler et al., "Machine Learning For Molecular And Materials Science", https://doi.org/10.1038/s41586-018-0337-2, pp. 547-555, Jul. 25, 2018.
Lin-Wang Wang, A special purpose computer for ab initio molecular dynamics simulations, Aug. 18, 2008, 30 Pages, Lawrence Berkeley National Laboratory, U.S. Website: https://escholarship.org/uc/item/2874t8zr.
Nigam, AkshatKumar, et al., "Augmenting Genetic Algorithms with Deep Neural Networks for Exploring the Chemical Space," ICLR 2020, arxiv.org, 2020, XP081579685, 14 pages.
Oleksii Prykhodko et al., "A De Novo Molecular Generation Method Using Latent Vector Based Generative Adversarial Network", Hit Discovery, Discovery Sciences, Biopharmaceutical R&D, AstraZeneca, & Department of Computer Science and Engineering, Gothenburg, Sweden, 27 pgs, Jun. 20, 2019.
Rafael Gómez-Bombarelli, et al., Automatic Chemical Design Using a Data-Driven Continuous Representation of Molecules, 2017, pp. 1-26, U.S.
Schmidt, Jonathan, et al., "Recent advances and applications of machine learning in solid-state materials science," NPJ Computational Materials, Aug. 2019, XP055733799, 36 pages.
THOMAS BLASCHKE; MARCUS OLIVECRONA; OLA ENGKVIST; J\"URGEN BAJORATH; HONGMING CHEN: "Application of generative autoencoder in de novo molecular design", ARXIV.ORG, CORNELL UNIVERSITY LIBRARY, 201 OLIN LIBRARY CORNELL UNIVERSITY ITHACA, NY 14853, 21 November 2017 (2017-11-21), 201 Olin Library Cornell University Ithaca, NY 14853 , XP080839053
U.S. Notice of Allowance dated Jun. 20, 2022, issued in U.S. Appl. No. 16/798,245 (9 pages).
U.S. Notice of Allowance dated Sep. 1, 2022, issued in U.S. Appl. No. 16/798,245 (5 pages).
Xiaolin Li et al. "A Deep Adversarial Learning Methodology For Designing Microstructural Material Systems", Proceedings of the ASME 2018 International Design Engineering Technical Conferences & Computers and Information in Engineering Conference IDETC/CIE, Aug. 26-29, 2018, 14pgs., Quebec City, Canada.
Yang, Zijiang, et al., "Microstructural Materials Design via Deep Adversarial Learning Methodology," arxiv.org, Cornell University Library, 2019, XP081015042, 34 pages.
YUAN DONG; DAWEI LI; CHI ZHANG; CHUHAN WU; HONG WANG; MING XIN; JIANLIN CHENG; JIAN LIN: "Inverse Structural Design of Graphene/Boron Nitride Hybrids by Regressional GAN", ARXIV.ORG, CORNELL UNIVERSITY LIBRARY, 201 OLIN LIBRARY CORNELL UNIVERSITY ITHACA, NY 14853, 21 August 2019 (2019-08-21), 201 Olin Library Cornell University Ithaca, NY 14853 , XP081466951
Zhaocheng Liu, et al., A Generative Model for Inverse Design of Metamaterials, 2018, pp. 1-15, Cornell University, U.S. Website: https://arxiv.org/abs/1805.10181.
ZIJIANG YANG; XIAOLIN LI; L. CATHERINE BRINSON; ALOK N. CHOUDHARY; WEI CHEN; ANKIT AGRAWAL: "Microstructural Materials Design via Deep Adversarial Learning Methodology", ARXIV.ORG, CORNELL UNIVERSITY LIBRARY, 201 OLIN LIBRARY CORNELL UNIVERSITY ITHACA, NY 14853, 8 May 2018 (2018-05-08), 201 Olin Library Cornell University Ithaca, NY 14853 , XP081015042, DOI: 10.1115/1.4041371

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20230409781A1 (en) * 2022-06-15 2023-12-21 Korea University Research & Business Foundation Inverse design system designing nano-optical device using controllable generative adversarial network and training and design methods
US11837333B1 (en) 2022-12-20 2023-12-05 Dow Global Technologies Llc Simulation guided inverse design for material formulations
US12587274B2 (en) 2023-03-28 2026-03-24 Quantum Generative Materials Llc Satellite optimization management system based on natural language input and artificial intelligence
US12368503B2 (en) 2023-12-27 2025-07-22 Quantum Generative Materials Llc Intent-based satellite transmit management based on preexisting historical location and machine learning
US12603701B2 (en) 2023-12-27 2026-04-14 Quantum Generative Materials Llc Distributed satellite constellation management and control system

Also Published As

Publication number Publication date
JP7733438B2 (ja) 2025-09-03
KR20210040248A (ko) 2021-04-13
EP3800586B1 (en) 2025-01-01
TW202117577A (zh) 2021-05-01
US20210103822A1 (en) 2021-04-08
CN112599208A (zh) 2021-04-02
EP3800586A1 (en) 2021-04-07
TWI846942B (zh) 2024-07-01
CN112599208B (zh) 2024-11-19
JP2021060992A (ja) 2021-04-15

Similar Documents

Publication Publication Date Title
US11537898B2 (en) Generative structure-property inverse computational co-design of materials
US20220383127A1 (en) Methods and systems for training a graph neural network using supervised contrastive learning
CN108108854B (zh) 城市路网链路预测方法、系统及存储介质
Shi et al. Effective decoding in graph auto-encoder using triadic closure
CN109919364A (zh) 基于自适应降噪和集成lstm的多变量时间序列预测方法
Gambs et al. Growing synthetic data through differentially-private vine copulas
CN112765415A (zh) 基于关系内容联合嵌入卷积神经网络的链路预测方法
CN116227562A (zh) 基于图神经网络与Transformer的时序点过程预测方法及系统
Díaz et al. Machine learning and statistical techniques. An application to the prediction of insolvency in Spanish non-life insurance companies
Liu et al. Multi-scale temporal-spatial feature-based hybrid deep neural network for remaining useful life prediction of aero-engine
Hashemi Deep generative models for ultra-high granularity particle physics detector simulation: A voyage from emulation to extrapolation
CN117763393A (zh) 一种基于神经网络特征选择增强的数据分类方法及装置
Cádiz-Leyton et al. Uncertainty estimation for time series classification-Exploring predictive uncertainty in transformer-based models for variable stars
Gkillas et al. Resource efficient federated learning for deep anomaly detection in industrial IoT applications
Benamira et al. Neural network-based rule models with truth tables
CN121329426A (zh) 交易风险识别方法、装置、电子设备、介质和程序产品
CN120045762B (zh) 一种基于边界节点条件gan的不平衡图节点分类方法及系统
Sima et al. MSO‐DETR: Metric space optimization for few‐shot object detection
Liu et al. SuperPruner: Automatic neural network pruning via super network
Stypułkowski et al. Representing point clouds with generative conditional invertible flow networks
Baek et al. A filter-and-refine approach to lightweight application traffic classification
Miao Precision-Recall Curve (PRC) Classification Trees (& Related Methods)
Hsieh et al. Housing price prediction by using generative adversarial networks
Wu et al. Feature fusion over hyperbolic graph convolution networks for video summarisation
John et al. Diverse GNN Encoder-Decoder for Graph Anomaly Detection

Legal Events

Date Code Title Description
FEPP Fee payment procedure

Free format text: ENTITY STATUS SET TO UNDISCOUNTED (ORIGINAL EVENT CODE: BIG.); ENTITY STATUS OF PATENT OWNER: LARGE ENTITY

STPP Information on status: patent application and granting procedure in general

Free format text: APPLICATION DISPATCHED FROM PREEXAM, NOT YET DOCKETED

AS Assignment

Owner name: SAMSUNG ELECTRONICS CO., LTD., KOREA, REPUBLIC OF

Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:HEGDE, GANESH;SIMKA, HARSONO S.;REEL/FRAME:052880/0319

Effective date: 20200221

STPP Information on status: patent application and granting procedure in general

Free format text: DOCKETED NEW CASE - READY FOR EXAMINATION

STPP Information on status: patent application and granting procedure in general

Free format text: NOTICE OF ALLOWANCE MAILED -- APPLICATION RECEIVED IN OFFICE OF PUBLICATIONS

STPP Information on status: patent application and granting procedure in general

Free format text: NOTICE OF ALLOWANCE MAILED -- APPLICATION RECEIVED IN OFFICE OF PUBLICATIONS

STPP Information on status: patent application and granting procedure in general

Free format text: NOTICE OF ALLOWANCE MAILED -- APPLICATION RECEIVED IN OFFICE OF PUBLICATIONS

STPP Information on status: patent application and granting procedure in general

Free format text: PUBLICATIONS -- ISSUE FEE PAYMENT VERIFIED

STCF Information on status: patent grant

Free format text: PATENTED CASE